Metadata-Version: 2.4
Name: jammi-ai-native-cu12
Version: 0.54.0
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Operating System :: POSIX :: Linux
Classifier: Environment :: GPU :: NVIDIA CUDA :: 12
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Rust
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Database
Requires-Dist: nvidia-cuda-runtime-cu12==12.6.*
Requires-Dist: nvidia-cublas-cu12==12.6.*
Requires-Dist: nvidia-curand-cu12==10.3.7.*
Requires-Dist: nvidia-cuda-nvrtc-cu12==12.6.*
Requires-Dist: nvidia-nccl-cu12==2.23.4.*
Summary: CUDA (cu12, sm_80+) build of the compiled in-process Jammi engine — the `jammi_native` PyO3 extension
Keywords: embeddings,vector-search,fine-tuning,inference,cuda,gpu
Author: F-Inverse
License: Apache-2.0
Requires-Python: >=3.9
Description-Content-Type: text/markdown; charset=UTF-8; variant=GFM
Project-URL: Bug Tracker, https://github.com/f-inverse/jammi-ai/issues
Project-URL: Documentation, https://f-inverse.github.io/jammi-ai/
Project-URL: Homepage, https://github.com/f-inverse/jammi-ai
Project-URL: Repository, https://github.com/f-inverse/jammi-ai

# jammi-ai-native-cu12

The compiled, in-process Jammi engine built for NVIDIA GPUs — the `jammi_native`
PyO3 extension with CUDA and the fused attention kernels, for compute capability
8.0 and newer (A100, A10, L4, L40S, H100, RTX 30/40-series).

```bash
pip install jammi-ai jammi-ai-native-cu12
```

```python
import jammi

db = jammi.connect("file:///data")   # the embedded engine, on the GPU
```

Install this **or** [`jammi-ai-native`](https://pypi.org/project/jammi-ai-native/),
the CPU build — both provide the `jammi_native` module. The CUDA libraries the
engine links come from NVIDIA's `nvidia-*-cu12` wheels, which this package
depends on; the host needs only an NVIDIA driver. Choose the device with
`[gpu] device` in the engine config; the CPU remains available on a machine with
no GPU.

The dist ships in lockstep at the shared workspace version with `jammi-ai`,
`jammi-ai-native` and `jammi-server`.

